narla.history

History

class narla.history.History(storage_size=10000)

Bases: object

History object contains arbitrary data from training

Parameters

storage_size (int) – Max storage size

clear()

Clear History

get(key, stack=False)

Access an element from the History

Parameters

• key (str) – Name of element to Access

• stack (bool) – If True will stack the values into a Tensor

Return type

list | torch.Tensor

record(**kwargs)

Store all the keyword arguments

Parameters

kwargs – If the key word doesn’t yet exist an internal list will be created and the value appended to it

sample(names, sample_size, from_most_recent=10000)

Return type

List[List[Tensor]]

stack(*keys, dim=- 1)

Stack a series of Tensors

Parameters

• keys – Keys of the values to access from the History

• dim (int) – Dimension of the Tensors to stack on

Return type

Tensor

to_data_frame(keys=())

Convert a History object to a DataFrame

Parameters

keys (List[str]) – Keys to use as columns in the DataFrame

Return type

DataFrame